Removing order effects from human-classified datasets: A machine learning method to improve decision making systems

Abstract

Although recent developments in Artificial Intelligence (AI) and machine learning (ML) aim to enhance the fairness and transparency of decision-making systems, research has found that neural networks (or other similar AI techniques) are still effected by human cognitive biases due to the training datasets. In this study, we focus on order effects, i.e., when the input of information impacts human perception and the decisions resulting from this information. We propose the Order Effect Removal Method (OERM) for handling the order effect which leads to bias and for helping organizations remove these biases from their training datasets and, therefore, from automated decision-making systems. Using design science principles to theoretically create, test, and validate the method, we can eliminate the order bias even in basic classification systems. Furthermore, the method can be applied in a multidisciplinary context, where an AI-based algorithm substitutes for manual work.